GitLab, PaperCut Exploited; AI Agents Aid Attacks

GitLab has been targeted by active exploitation of CVE-2026-85706, a critical path traversal vulnerability that allows unauthenticated users to read arbitrary files. This flaw affects all versions of GitLab CE/EE prior to specific patch releases in 19.1, 19.2, and 19.3. The vulnerability was added to CISA's Known Exploited Vulnerabilities (KEV) catalog, highlighting its immediate threat. Multiple news outlets, including GovInfoSecurity and The Register, reported on the active exploitation occurring shortly after the patch was released, underscoring the urgency for organizations to update their GitLab instances. Rapid7 has also published analysis and Metasploit modules for this vulnerability.

Exploitation of critical vulnerabilities in PaperCut MF and NG, specifically CVE-2026-81578 (improper access control) and CVE-2026-82078 (unsafe dynamic class loading), has been observed in the wild. These flaws allow unauthenticated remote attackers to execute arbitrary code and gain control of affected systems. Reports indicate that attackers have leveraged hundreds of AI agents in these campaigns, compromising over 440 instances, with a particular focus on educational institutions. CISA has also issued warnings about these actively exploited vulnerabilities, emphasizing the need for prompt patching. The Hacker News and SecurityWeek have extensively covered the escalating exploitation and the methods used by attackers.

CVE-2026-25470, a critical unauthenticated remote code execution vulnerability in the ACPT (Pro) - Custom Post Types Plugin for WordPress, poses a significant risk to WordPress sites. Versions up to 2.0.47 are affected, allowing attackers to execute arbitrary code on vulnerable servers. While specific details on active exploitation are emerging, the critical severity and unauthenticated nature of this flaw make it a prime target for malicious actors.

Apache Artemis and ActiveMQ Artemis are affected by CVE-2026-27446, a critical vulnerability that allows unauthenticated remote attackers to force a target broker to establish an outbound Core federation connection. This missing authentication for a critical function could lead to various security compromises depending on the network environment and broker configuration.

CVE-2026-90919, a critical remote code execution vulnerability in LightLLM through version 1.2.0, has been identified. The flaw resides in the Config Server's unauthenticated /visual_register WebSocket endpoint, which passes client data directly to pickle.loads() without proper validation. This allows attackers to execute arbitrary code on the server by sending a crafted payload.

CVE-2026-89094, a critical remote code execution vulnerability in Forgejo versions prior to 16.0.4, allows attackers to execute arbitrary code by leveraging a crafted template repository. The vulnerability stems from mishandled template expansion on files within the .forgejo/template directory, enabling attackers to inject malicious code.

CVE-2026-48273, a critical 'Eval Injection' vulnerability in Adobe ColdFusion, allows for arbitrary code execution in the context of the current user. This flaw can be exploited by low-privileged attackers, making it a significant threat to organizations using affected versions of ColdFusion. SecurityWeek has noted Adobe's patching of this and numerous other vulnerabilities.